Studies On The Leaf Information Of Rapeseed Based On The Computer Vision Technology

The shortcomings of obtaining leaf information by traditional method are time-consuming, laborious, Inaccurate and none real-time. In this paper, take the threedifferent types leaves of Brassica napus as the main object, we try to do some researchon the leaf area and color information base on the graphics theory and using computervision technology, so as to do some preliminary groundwork for the establishment ofthe nondestructive, real-time and accurate crop information obtaining system. Thestudy involves following works:(1) In the area information extraction of the three types of rapeseed leaf, with thepaper weight method as a standard, the accuracy of ruler and computer vision methodare contrasted. The results show that for the long and short petiole leaf, the averagerelative error of ruler method are 6.41% and 5.11%, but for computer vision methodthey are 4.16% and 2.48%, the latter one only has less advantage in accuracy. But forthe no petiole leaf, the average relative error of ruler method is 10.49% when thecoefficient is 0.7, and for the computer vision method, it is 3.09%, the advantage ofcomputer vision method in accuracy is obvious. According to the measured value ofstandard method, we correct the coefficient of ruler method for no petiole leaf andfound that 0.75 is appropriate.(2) Based on the computer vision technology, 27 color characteristics ofrapeseed leaves are extracted from the digital image and their difference result fromdifferent taking time under natural light conditions (morning, noon and evening) are alsoanalyzed. It shows that the relative between color characteristic and SPAD (the relativecontent of chlorophyll) is the best when the leaf image is taken at noon, comparing therelative between natural and indoor light conditions further, we found that the latter oneis better and the obverse side is more accurate than the reverse side. (3) The measuring models are established by regression analysis on the value ofSPAD and color characteristics. In the natural light conditions (at noon), the predictedaccuracy of the B/R, r-b and B/(R+G) model is over 90%, and in the indoor lightconditions, the accuracy of the B/G, b, S and B/(R+G) model is more higher, thedetermination coefficient of B/G forecast model is 0.9081.(4) The article discusses the accurate recognition and area measuring on holes,petioles and veins of the leaf based on the computer vision technology. The results laida foundation for the next pest spots analysis and information acquisition of rapeseed.These studies indicate that computer vision application in capturing theinformation of rapeseed leaves is feasible, and it is more effective when we needprecision leaf area and obtain SPAD value rapidly. The results provide a theoreticalbasis and a reasonable model for the nondestructive acquisition and real-timemonitoring of the crop information. It also laid a foundation for the research on portableequipment which can be used to obtain crop information rapidly.